Temporal-context 1D convolutional neural network for joint modulation and transmission-distance classification in visible light communication

Abstract

Accurate identification of modulation format and transmission distance is important for reliable visible light communication (VLC) systems, particularly under channel-dependent signal distortions. This study presents a temporal-context one-dimensional convolutional neural network (1D-CNN) for simultaneous modulation-format recognition and transmission-distance classification using experimentally acquired VLC signals. The proposed network employs four convolutional layers with batch normalisation and Leaky ReLU activation, followed by global average pooling, a fully connected layer, and two softmax output heads for modulation-format and transmission-distance classification. A signal segment length of N = 40 and a temporal-context of  K = 10 were selected based on comparative experiments. The framework was evaluated for seven modulation formats and 29 discrete transmission-distance classes from 0 to 140 cm at 5-cm intervals, achieving 99.5% modulation-format classification accuracy and 98.7% distance-classification accuracy. The distance results demonstrate discrete identification of experimentally measured link positions rather than general-purpose distance ranging. The results further show that classification performance decreases for higher-order modulation formats and longer transmission distances because of increased constellation density and channel distortion. The proposed approach provides a compact solution for joint signal-format and distance identification directly from received temporal waveforms.